aksonocr1.5 · Document Intelligence API

Turn documents into structured, spatial data.

Parse PDFs and images into Markdown, HTML, reading-order elements, tables, figure descriptions, and normalized polygon coordinates. Built for production workflows with asynchronous jobs, idempotency, and signed webhooks.

Start integrating Get an API key

Quickstart

Submit once, then poll the canonical status URL until the request reaches a terminal state.

curl --request POST \
  --url https://backend.aksonocr.com/api/v1/document-parse \
  --header 'X-API-Key: <YOUR_API_KEY>' \
  --header 'Idempotency-Key: invoice-2026-001' \
  --form '[email protected]' \
  --form 'mode=accurate' \
  --form 'include_text_lines=true' \
  --form 'parse_charts=true' \
  --form 'parse_images=true'

Submit a document

POST /api/v1/document-parse · multipart/form-data · returns HTTP 202 Accepted

Field	Type	Required	Description
document	file	Conditional	PDF, PNG, JPEG, or WebP; maximum 50 MB.
document_url	string	Conditional	Public HTTPS URL. Use instead of document.
mode	enum	No	fast or accurate (default).
page_start / page_end	integer	No	Inclusive 1-indexed PDF page range.
include_text_lines	boolean	No	Include fine-grained OCR line elements. Default true.
include_page_images	boolean	No	Embed page previews for overlays. Default false.
parse_charts	boolean	No	Parse detected chart regions into Markdown tables. Default true.
parse_images	boolean	No	Describe image, figure, and header_image regions with a VLM (crop kept). Default true.
webhook_url	URL	No	HTTPS endpoint for terminal-state notifications.

document and document_url are mutually exclusive. Send an Idempotency-Key header when retrying submission requests.

Request lifecycle

Every request is a resource. Read the same status_url for progress and the final inline result.

pending

processing

completed

failed

cancelled

{
  "api_version": "2026-08-01",
  "request_id": "req_01...",
  "status": "pending",
  "model": "aksonocr1.5",
  "status_url": "https://backend.aksonocr.com/api/v1/document-parse/req_01...",
  "progress": 0,
  "created_at": "2026-08-01T08:00:00.000Z",
  "expires_at": "2026-08-04T08:00:00.000Z"
}

Result schema

Completed requests add result to the job object. Content is document-level; elements provide page-level structure and geometry.

{
  "api_version": "2026-08-01",
  "request_id": "req_01...",
  "model": "aksonocr1.5",
  "mode": "accurate",
  "document": { "filename": "invoice.pdf", "total_pages": 1 },
  "usage": { "pages_processed": 1, "credits_used": 0.001 },
  "content": { "html": "...", "markdown": "...", "text": "..." },
  "pages": [{ "page": 1, "width": 1654, "height": 2339 }],
  "elements": [{
    "id": "element_0",
    "type": "table",
    "page": 1,
    "parent_id": null,
    "reading_order": 3,
    "confidence": 0.97,
    "polygon": [
      { "x": 0.08, "y": 0.31 }, { "x": 0.92, "y": 0.31 },
      { "x": 0.92, "y": 0.67 }, { "x": 0.08, "y": 0.67 }
    ],
    "bounding_box": { "x": 0.08, "y": 0.31, "width": 0.84, "height": 0.36 },
    "content": { "html": "<table>...</table>", "markdown": "| ... |", "text": "..." }
  }]
}

Coordinates

polygon is the source of truth and may contain more than four points for dewarped pages. bounding_box is derived for simpler clients.

Normalized space

Both axes range from 0 to 1. Multiply x by the rendered page width and y by its height.

Stable orientation

Points describe the detected outline in page coordinates. Do not assume every polygon is axis-aligned.

const points = element.polygon
  .map(({ x, y }) => `${x * imageWidth},${y * imageHeight}`)
  .join(" ");

Webhooks

Terminal events are sent as document_parse.completed or document_parse.failed.

When a webhook secret is configured, requests include X-Akson-Signature: sha256=<digest>. Verify the HMAC-SHA256 digest against the raw request body before processing the event.

Errors

Use HTTP status codes for transport errors and error.code for programmatic handling.

HTTP	Code	Description
400	invalid_request	Invalid options, document, or URL.
401	unauthorized	Missing or invalid API key.
404	not_found	The request does not exist or is not owned by this key.
429	rate_limited	Retry with exponential backoff.
503	service_busy	GPU capacity is temporarily unavailable.